Repository navigation
Conversation
bdice
added
bug
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Contributor
Author
|
/ok to test abccc02 |
Contributor
Author
|
/ok to test 688c279 |
Contributor
Author
|
/ok to test 594be17 |
Contributor
Author
|
/ok to test 1e5291c |
Contributor
Author
|
/ok to test 58a0031 |
Contributor
Author
|
/ok to test 7775653 |
Contributor
Author
|
/ok to test aa6cd2d |
Contributor
Author
|
/ok to test a2f8397 |
Contributor
Author
|
/ok to test 3bf411f |
Contributor
Author
|
/ok to test bde6679 |
1 similar comment
Contributor
Author
|
/ok to test bde6679 |
Contributor
Author
|
/ok to test 018c20e |
Contributor
Author
|
/ok to test e0eccc4 |
Contributor
Author
|
Closing: the crash is a UCX 1.22.0 bug (SIGSEGV in cuMemGetHandleForAddressRange when UCX probes dmabuf export on async managed pool memory), fixed on UCX main by openucx/ucx#11922 and tracked in rapidsai/ucxx#749. rapidsai/ucxx#784 excludes ucx==1.22.0, which will reach cudf CI through ucxx. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Draft for diagnosing intermittent failures of two conda jobs on
13.3.0, 3.13, arm64, ubuntu26.04, l4, latest-driver, latest-depsseen on #24663:conda-python-cudf-polars-tests: hits the 30 minute PR timeout (exit 124) after a RayRankActordies.conda-python-other-tests:signal 11incudf_streaming/tests/test_allgather.py, also seen on Convert quantiles' and reductions' scratch RMM vectors to CUDA buffers #24638.Evidence from the #24663 polars log (run 39198377175): a
RankActortook a SIGSEGV incuMemGetHandleForAddressRangecalled fromuct_cuda_copy_md_get_dmabuf->ucp_memory_detect_slowpath->ucp_tag_recv_nbx->rapidsmpf::coll::AllGather::event_loop. The surviving rank then waits forever in the all-gather and the engine teardown (ray.getwith no timeout) blocks until the job is killed.This PR only builds and tests that configuration (three copies of each test job to measure how often it fails). It must not be merged in this form.